Health impacts of the Cambridgeshire Guided Busway: a natural experimental study

Cohort study

Intercept survey

The aims of the intercept survey were to obtain a random and more socioeconomically heterogeneous sample of busway users, which would enable comparisons with the main study cohort and partly offset the volunteer bias in recruitment to the cohort, and to study the perceptions of the busway and its impact on travel behaviour in this sample.

Qualitative investigations

We also explored the commuting contexts, decisions and experiences of participants in a series of qualitative substudies. This entailed more than 120 semistructured interviews, as well as a photo-elicitation study, ethnographic participant observation, media analysis and a qualitative study of the final stakeholder forum for the project.

Incentives and feedback for participants

Participants were entered into a prize draw to win one of eight £50 gift vouchers as an incentive to participate in each phase of the cohort study and in the intercept survey. The study team produced an annual newsletter to inform participants of the emerging study findings. In addition, those who completed basic activity monitoring were offered brief individualised feedback in the form of a bar chart comparing their recorded daily minutes of moderate to vigorous physical activity (MVPA) with the average for their age group and gender in the sample, and with current physical activity guidelines. Those who completed data collection with combined heart rate and movement sensors were offered more detailed individualised feedback on their activity data.

Active commuting

The primary outcome for the study was the time participants spent in active commuting, that is the time they spent walking or cycling to and from work each week.

Mode of travel to work

We used information on mode of travel to work from the core questionnaire to summarise and categorise the travel behaviour of our sample in other ways.

Use of the busway

Use of the busway was summarised using three binary variables indicating whether participants had ever used the guided bus service, ever used the path for walking or ever used the path for cycling. These were derived from the responses to two questions in the core questionnaire: ‘Have you ever travelled on a guided bus in Cambridgeshire?’ (response options: ‘yes’ or ‘no’) and ‘Have you ever walked or cycled along any part of the footpath or cycle path beside the guided busway?’ (response options, of which more than one could be selected: ‘yes – I have walked beside the busway’, ‘yes – I have cycled beside the busway,’ or ‘no – I have not walked or cycled along the path beside the busway at all’).

Physical activity

Estimates of time spent in MVPA were derived either from the information in the core questionnaire (‘reported’) or from objective measurement (‘recorded’), while estimates of PAEE were derived from the combined heart rate and movement sensor data.

Carbon emissions

We also investigated the impact of the busway on changes in carbon dioxide emissions attributable to commuting in our cohort between 2009 and 2012. We estimated the carbon emissions for each commuting trip reported by each participant and summed these across the week to produce a total for each participant at each time point. Carbon emissions were estimated by multiplying the distance travelled by each mode of transport by a published emissions factor specific to each mode of transport, and specific to each vehicle where possible. 59

The distance travelled on each trip to and from work was estimated using the Google Directions API (application programming interface) (Google™, Google Inc., Mountain View, CA, USA) with the origin and destination of the trip. These were derived from unit postcodes but shifted by 1 minute of longitude or latitude to prevent the release of potentially identifiable pairs of information (i.e. linked home and workplace locations) to Google. Distance calculations took into account the mode or modes of transport used. For participants who reported a park-and-ride commute, for example, the trip was divided into two parts, and the Google option for mode of transport was set to ‘driving’ for the first part of the trip and ‘transit’ (i.e. public transport) for the second.

For guided bus trips, we used the emission factor for trams (87.61 gCO₂/km), on the basis that the environmental performance of a guided bus is broadly comparable with that of light rail, and that if the greater flexibility for timetable optimisation is taken into account a guided bus system can match the carbon emissions of trams. 60 For bus trips, we also estimated the speed for each trip based on the distance travelled, time travelled and number of stops. Trips with an average speed of > 50 miles per hour were defined as ‘coach’ trips, and others as ‘local bus’ trips, and emission factors were assigned accordingly. For car travel, we drew on the information from our core questionnaire to distinguish between fuel type (petrol, diesel or hybrid) and engine size (< 1.4 l, 1.4–2.0 l or > 2.0 l) of cars used at each time point to apply the appropriate emission factors.

Indicators of health and well-being

Exposure to the intervention

Evaluation studies, especially in health research, are often carried out by comparing outcomes in intervention and control groups, preferably created by the random allocation of individuals or groups to the treatment conditions. In evaluations of place-based interventions such as new transport infrastructure, however, randomisation is generally not possible and it is not necessarily meaningful to treat exposure to the intervention as a binary variable. In the light of this, a number of approaches for modelling ‘exposure’ in quasi-experimental studies of place-based interventions have been developed. Exposure may be modelled in terms of coincidence or adjacency, for example by denoting a particular administrative area or a buffer zone the size of a ‘walkable’ area around the intervention. Alternatively, exposure may be modelled in terms of varying intensities, either ordinal or continuous, derived from the proximity of a given participant’s location to the intervention. Measures of exposure may be further refined by using additional information on the characteristics of individual participants, such as their travel patterns.

Exploratory analysis of the characteristics of our cohort showed that comparable intervention and control groups could not be created based on area of residence. For example, the distribution of participants who reported incorporating walking or cycling in their commute at baseline appeared neither random nor associated in any singular way with the planned route of the busway. Participants living closer to the planned route were more likely to report cycling for commuting, but less likely to report walking for commuting, than those living further away (Figure 2). This can be explained in part by the fact that distance from the busway was correlated with distance from the city centre and that local commuting patterns were inevitably reflected in our achieved sample (see Chapter 4, Patterns of active commuting). For example, the tendency of commuters to combine walking with using a car or public transport meant that reporting some walking for commuting remained relatively common even among those living furthest from Cambridge, while the prevalence of cycling for commuting declined sharply at distances > 10 km. 62 Similar location effects could be seen in other characteristics of our sample, for example in the tendency for younger and more mobile participants to live closer to the planned route of the busway.

FIGURE 2.

Distribution of active commuting at baseline by proximity to the route of the busway. Proportion of participants within each distance interval reporting any walking or cycling to and from work at baseline (for walking, n = 456; for cycling, n = 468).

Rather than assigning participants to simple ‘intervention’ and ‘control’ areas, which would have been unbalanced on the baseline distribution of active commuting and other socioeconomic covariates, we therefore computed graded, participant-level measures of intervention exposure. Our graded measure of exposure was based on the distance from each participant’s home to the nearest access point to the busway. We calculated this distance using route networks created in the GIS (see Derivation of key variables).

In studying the predictors of using the busway for a given mode of transport (bus, walking or cycling) the graded measure of exposure was adapted to take account of the mode of transport in question (see Chapter 5, Patterns and predictors of busway use). For the outcome of guided bus use, we computed the distance from each participant’s home to the nearest busway stop using any combination of the road and pedestrian and cyclist route networks (including the path along the busway). This reflected the fact that, in order to travel on a guided bus, participants would first have to travel to a bus stop using either a car or an active mode of transport, and in the latter case could use the path along the busway to do so. For the outcomes of walking or cycling on the path, we computed the distance from each participant’s home to the nearest access point to the path, using the same route networks.

In studying the effect of the busway on time spent in active commuting and physical activity, we hypothesised that the busway could promote walking and cycling either as modes of travel along the path or as feeder modes to the guided bus service (see Chapter 5, Linking environmental change with travel behaviour change in commuters). We therefore used the distance from each participant’s home to either the nearest busway stop or the nearest access point to the path, whichever was closer.

Exploratory analysis further showed that a given increment in distance had a smaller effect on use of the busway as distance increased, which suggested that it would be inappropriate to model exposure to the intervention as a linear function of distance. Although the log transformation is commonly used in studies of distance decay in transport studies, we selected a square root transformation because the estimates produced were slightly more conservative and more easily interpretable. For ease of interpretation, we defined exposure to the intervention in terms of ‘proximity’ rather than distance, proximity being defined as the negative square root of the distance. For example, for a pair of participants living 9 km and 4 km, respectively, from the busway, the measures of proximity were –2 and –1, respectively. Because –1 is larger than –2, this enabled us to report the effects of ‘increasing proximity’ to the busway.

In studying the predictors of using the busway for a given mode of transport, we also explored the application of more detailed participant-level measures of intervention exposure (see Chapter 5, Predictors of busway use). We hypothesised that individuals would be more ‘exposed’ to the intervention if the new infrastructure were to reduce the time taken to travel to work. This reduction in travel time could result from a reduction in either the bus journey time, or the walking or cycling distance from home to work, and would vary according to the work location as well as the home location of the participant. For example, a participant living close to the busway and working at a location close to the route might experience a reduction in the time it would take to walk or cycle to work, whereas a neighbour living next door but working at a location further from the busway might not benefit to the same extent. These ‘individually calibrated’ measures of intervention exposure were calculated using each participant’s home and workplace addresses. We computed changes in estimated travel time to work by bus using Accession, a software package that uses road networks, road speed data and public transport timetable information to calculate travel times (Citilabs Inc., Lafayette, CA, USA). We used this to estimate the quickest potential bus journey for each participant on the networks that existed before and after the completion of the busway on a Wednesday, with the requirement that the journey should commence after 06.00 and end before 10.00 to correspond to a typical morning commute. The change in the modelled walking or cycling distance to work induced by the busway was calculated using the GIS by comparing the distances in the pedestrian and cyclist route network before and after addition of the path along the busway.

Other environmental variables

We also studied the influence of subjective and objective factors on active commuting. Our core questionnaire contained questions on participants’ perceptions of the environment on their route to work. We also used the GIS to derive variables corresponding to objective features of the environment.

Psychological measures

Sample size estimation

In our original study design we aimed to achieve a sample of 788 participants at 1-year follow-up, which would have provided 80% power to detect a standardised mean difference between intervention and control groups of 0.2, equivalent to a mean increase of 2 minutes per day in our primary outcome of time spent in active commuting. 32 The study was not designed to have comparable statistical power to detect changes in the secondary outcomes. The delayed completion of the busway resulted in an enforced follow-up period of 3 years rather than 1 year and a concomitant reduction in the final sample size for analysis. This smaller achieved sample would have provided 80% power to detect a mean difference of 4 minutes per day, which, although larger than the original target effect size, is still commensurate with the effect sizes observed for comparable interventions in previous studies. 27,69 In the event, the distribution of the primary outcome data precluded the use of a continuous outcome measure to model the effect of the intervention (see Chapter 5, Linking environmental change with travel behaviour change in commuters).

Quantitative analysis

Many different quantitative analyses are summarised in this report. Further details of specific analyses can be found in subsequent chapters and in the relevant publications, but our general approach was to use multivariable regression modelling to estimate adjusted associations between dependent (‘outcome’) and independent (‘exposure’) variables. Most analyses involved the use of linear, logistic or multinomial regression to model continuous, binary and categorical outcomes, respectively. We built up these models in stages by progressively adjusting them for various sets of individual, household, geographic and other covariates that were hypothesised to be potential confounders of the relationships of interest. We generally selected such covariates to be taken forward into multivariable models if they met the generous statistical criterion of being associated with the dependent variable at p < 0.25 in univariable analysis. We stratified some models by prespecified effect modifiers, for example by gender in the analysis of the correlates of physical activity and by baseline commuting behaviour in the main outcome analyses. We mostly did not impute missing data for covariates, and we never imputed missing data for behavioural or health outcome variables.

Qualitative analysis

The qualitative substudies produced a range of qualitative data, most notably interview transcripts but also participant-produced photos, ethnographic field notes, and documents (newspaper articles and tweets). All qualitative data were analysed using variations of thematic content analysis in which data were categorised and synthesised into emerging themes. The initial coding was usually carried out by the researcher who collected the data in the context of a particular substudy, with subsequent refinement and interpretation in collaboration with other members of the research team. Various theoretical frameworks informed particular analyses, for example practice theory and normalisation process theory from the fields of medical sociology and anthropology, respectively (see Chapter 4, Active commuting in context, and Chapter 5, Understanding the intervention).

Rationale

Accurate measurement of active commuting and associated physical activity in free-living conditions is important for describing patterns in these behaviours and studying their determinants in populations. It is also essential if we are to have confidence in the effect sizes from intervention studies and for the attribution of subsequent health impacts. A validated, short, self-reported measure of time spent walking or cycling for transport was not available when data collection for our study began in 2009. Many studies have used comparatively simple measures of travel behaviour. These have often focused on either the main mode of transport, that is the mode used for the longest part of a trip (if more than one mode is used), or on the usual mode of transport, that is the mode used most frequently over a specified period of time. Neither approach is entirely satisfactory for estimating the physical activity associated with commuting because each entails simplifying the detail and complexity of travel behaviour. Focusing on the main mode of transport means that active travel may be underascertained, because actual commuting journeys, especially those involving public transport, often involve more than one mode of transport (‘multimodal’ trips) and may involve walking or cycling at either end of the trip. Focusing on the usual mode of transport, on the other hand, may fail to capture day-to-day variation in the mode, distance or duration of trips. Although some validation of the modes and frequency of trips recorded in detailed day-by-day travel diaries has been reported, it remains important to validate measures of time spent in active commuting because this information can be used to quantify the health impacts of commuting. 70

The selection of a ‘gold standard’ method for validating self-reported measures in this field is challenging. Direct observation of participants, as well as other methods such as wearable cameras, have been used in other, smaller studies, but it is unlikely that these would have been acceptable to participants or practical to implement in a study of our size. 71 Likewise, the identification of walking and cycling from the data captured by physical activity monitors is not sufficiently advanced to enable the time spent in active travel to be confidently determined. 72 A similar problem exists with the use of GPS data alone. These have been used to infer mode of transport from either the maximum or the average speed recorded in a trip, whereas in our study many participants commuted into city centre locations at low average speeds, often using a combination of modes of transport, which would have increased the risk of misclassification. 73,74

We therefore developed a new method for obtaining objective estimates of the time spent using active and motorised modes of transport, by combining reported data on mode of transport with georeferenced combined heart rate and movement sensor data in a subsample of our study cohort. We then used these objective estimates to assess the convergent validity of self-reported estimates of time spent in active commuting derived from our core questionnaire data. 75

Methods

In order to generate objective estimates of time spent using active and motorised modes of travel, and total commuting travel time, it was first necessary to process the GPS data to identify commuting trips and stages within trips. Because the processing and cleaning of trip-level GPS data is a labour-intensive task, we selected a quasi-random sample of commuters from the larger subsample who provided combined heart rate and movement sensor and GPS data, as well as information about their commuting trips in the past week in both the travel diary and the core questionnaire, in either 2010 or 2011. We aimed to include at least 50 commuting trips for each of six combinations of mode of transport found in the cohort study: (1) walking, (2) cycling, (3) car or motorcycle only, (4) car in combination with walking, (5) car in combination with cycling and (6) bus with or without walking or cycling (see Chapter 4, Patterns of active commuting). Because relatively few participants had recorded walking all the way to or from work, reflecting our recruitment strategy and local commuting patterns, we achieved only 42 such trips for analysis.

We used the GIS to project the GPS data onto mapping information and to locate the home and workplace of each participant. We then inspected the GPS tracks to obtain an objective estimate of the total duration of each commuting trip by calculating the difference between the time at which participants left home or entered the building outline of their workplace. If the trip involved only one mode of transport (a ‘unimodal’ trip), it was added to the data set for analysis at this stage.

If a trip involved a combination of active and motorised travel, we needed to obtain objective estimates of the duration of each stage of the trip. For trips that combined walking or cycling with car travel, we were able to identify the time at which an active stage of the trip began from the combined heart rate and movement sensor data alone, but this proved impossible for trips that involved walking or cycling in combination with bus travel. We therefore cross-referenced the GPS data with the locations of bus stops, which enabled us to more clearly identify the start and end times of the bus stages of trips. Waiting times at bus stops before or after boarding were identified by GPS points with non-regular speeds in close geographic proximity. These waiting periods were excluded from analysis, along with any periods of longer than 5 minutes spent at a location en route (such as a school or shop) if no change in mode of transport was reported.

Once the different stages of commuting trips had been identified, we calculated our objective estimates of time spent using active and motorised modes of travel, as well as the total duration of the commuting trip. We then carried out three comparisons to assess the convergent validity of our self-reported measures of active commuting:

1. We compared estimates of the duration of commuting trips, and stages of trips, spent using active and motorised modes of transport, as well as total trip duration, all derived from the detailed travel diary, with the corresponding objective estimates.

2. We compared estimates of usual time spent using active modes of transport derived from the core questionnaire with the corresponding objective estimates.

3. We compared the two self-reported estimates, namely the estimates of usual time spent using active modes of transport derived from the core questionnaire with those derived from the detailed travel diary.

If data were of insufficient quality to allow modal transitions and destinations en route to be confidently identified, trips were included only in the analysis of total trip duration and excluded from the analyses of time spent using active or motorised modes.

Results

We carried out these comparisons separately for commuting trips that involved different combinations of modes of transport. In the first comparison these showed that in the travel diary, participants overestimated the time they spent walking or cycling to work by 1.13 minutes per trip on average [95% limit of agreement (LOA) −7.67 to 9.95; p = 0.001]. The median duration of these trips was 38 minutes. The mean overestimation of active travel time was slightly larger in the subset of trips made by walking or cycling alone (2.39 minutes per trip, 95% LOA −6.78 to 11.55), even though these trips tended to be shorter, with a median duration of 17 minutes. However, the difference between objective and self-reported estimates in this subset of trips was not significant (p = 0.25). For the subset of trips involving active travel in combination with any motorised mode of transport, the magnitude of the difference was smaller (mean +0.40 minutes per trip) but statistically significant (p = 0.001) (Table 5).

As we discuss in more detail below (see Interpretation), this suggests that participants are able to provide acceptably accurate estimates of their active travel time for trips that also involve other modes of transport in a travel diary. In contrast, we found that participants’ estimates of time spent using motorised modes of travel were subject to a substantially larger overestimation than were their estimates of active travel time, with an average overestimation of 10.48 minutes per trip (95% LOA −13.62 to 34.58). Trips involving motorised modes in combination with active modes appear to be responsible for this finding (mean difference 14.49 minutes per trip, 95% LOA −8.86 to 37.84; median total trip duration 45 minutes).

When we compared estimates of usual active commuting time derived from the core questionnaire with the corresponding objective estimates, we found differences of a similar magnitude to those observed in the travel diary comparisons. The difference was smaller for cycling, with an average underestimation of 1.12 minutes per trip compared with a median reported usual time of 17 minutes (95% LOA –8.67 to 6.44, n = 11 participants), than for walking, with an average overestimation of 2.37 minutes per trip compared with a median reported usual time of 13 minutes (95% LOA –10.91 to 15.64, n = 16 participants); neither difference was statistically significant (p > 0.35). The questionnaire and objective measures showed stronger agreement for time spent cycling (with a concordance coefficient of 0.96) than for walking (with a concordance coefficient of 0.84).

In our third comparison, we found that the questionnaire estimates of usual active travel time were not significantly different from the mean duration derived from the trips reported in the travel diary (p > 0.1), with an average difference of less than 1 minute per trip for both cycling and walking.

Interpretation

When we compared estimates of active and total travel time from day-by-day travel diaries and from retrospective questionnaires to objective estimates, both self-reported measures performed acceptably well for our purposes. The mean differences between self-reported and objective estimates were relatively small, both for the detailed travel diary (an overestimation of approximately 1 minute per trip) and for the retrospective travel record. These findings are similar to those of previous validation studies. 71 The results of the travel diary comparison further suggest that where trips involved a combination of active and motorised modes of transport, participants tended to report the duration of the active stages more accurately than the duration of the motorised stages. The small mean overestimation in the self-reported measures indicates that such methods can provide a reasonably accurate estimate of aggregate active commuting time in a population. Nevertheless, the wide LOAs indicate a risk of substantial over- or underestimation at an individual level, which means that self-reported measures should be used with caution to infer aggregate weekly quantities of physical activity as part of commuting. For example, a 2-minute overestimation per trip corresponds to a 20-minute difference over a 5-day working week. The effect sizes of the most promising interventions to promote walking for transport are of the order of 15–30 minutes per week, which is directly comparable with the size of the measurement error suggested by our analysis. 27 This suggests that evaluations of interventions that are anticipated to have relatively modest effect sizes may have limited power to detect effects on time spent in active commuting if they rely on self-reported measures alone, because the potentially large measurement error may mask small but important real changes in behaviour.

Rationale

Although the measurement of physical activity is an area of investigation in its own right and our study was not designed to compare different measures of physical activity, we explored the possible consequences of using different measures of physical activity in two papers.

In the first paper, we investigated which of a range of sociodemographic, health and geographic characteristics were associated with either self-reported (‘reported’) or accelerometer-measured (‘recorded’) estimates of daily time spent in MVPA. 76 We did this because our knowledge about the extent to which the correlates of physical activity vary according to the approach to measurement is limited. 77 If different measures were to identify different correlates, this would in turn suggest different determinants of change in physical activity and the possibility that different intervention effects might be detected with different measurement methods. In the second paper, we set out to understand the characteristics of those whose physical activity level was ranked differently according to the approach to measurement. 78 This is important because such mismatches may complicate the assessment of the effects of interventions.

Methods

Both papers used baseline cohort data from 2009. In the first paper, ‘recorded’ daily time in MVPA was derived as previously described (see Chapter 2, Derivation of key variables). 76 In the second paper, we processed the accelerometer data to obtain average daily accelerometer counts per minute for each participant. 78 Participants were further assigned to one of three groups according to whether their average level of physical activity was (1) ranked equally (i.e. categorised in the same quartile) by self-report and accelerometer, (2) ranked higher (i.e. categorised in a higher quartile) by self-report than by accelerometer, or (3) ranked higher by accelerometer than by self-report. For both papers, participants needed to provide accelerometer data on at least 3 days to be included in analysis, and the analysis was carried out separately for men and women because previous research suggests that the correlates of physical activity differ by gender. Multiple regression models of recorded daily time in MVPA were adjusted for accelerometer wear time, because participants who wore the accelerometer for longer periods tended to accumulate more minutes of physical activity.

Results

The first paper showed that different correlates were associated with ‘reported’ and ‘recorded’ daily time spent in MVPA, and few variables were associated with both measures in either men or women. 76

In men, four individual characteristics were significantly associated with reported MVPA. Men who had a standing or manual occupation as opposed to a sedentary occupation, who did not have at least degree-level education, who had access to a bicycle and who reported a mental component summary score above the lowest quartile reported substantially more MVPA. Different variables were significantly associated with men’s recorded MVPA. Men who lived with additional adults in the same household, or in areas with higher proportions of greenspace, achieved lower levels of recorded MVPA.

In women, three variables were significantly associated with reported MVPA. Women with access to a bicycle and with degree-level education reported more time spent in MVPA per day, while women in households with at least one child reported less. In contrast, one variable – having at least degree-level education – was associated with higher recorded MVPA. Both having access to a car and being overweight or obese were associated with lower recorded MVPA. Although the directions of associations tended to be consistent across both measures for variables such as age, access to a bicycle and weight status, especially for women, not all of these associations were statistically significant.

In the second paper, we found that approximately one-third of participants were ranked in the same quartile with both physical activity measures, while one-third were ranked lower, and one-third ranked higher, on one measure than the other. 78 In adjusted analyses, the physical activity of overweight or obese individuals was less likely to be ranked in a higher quartile by an accelerometer than that of normal-weight individuals [odds ratio (OR) 0.54, 95% confidence interval (CI) 0.32 to 0.92] and twice as likely as that of normal-weight individuals to be ranked in a higher quartile by self-report (OR 2.07, 95% CI 1.28 to 3.34). These overall results were replicated in the separate analysis for women, but not in that for men. This suggests that individual characteristics such as gender and weight status may be associated with mismatches in estimates of physical activity between different measures.

Both analyses included a smaller number of men than women, which may have further reduced the power to detect associations in this group. In addition, the estimates of reported and recorded MVPA used in these analyses were neither synchronous nor of matching duration, in that ‘recorded’ activity was measured over 7 days, whereas ‘reported’ activity was captured in terms of weekly frequency and average duration over the past 4 weeks and transformed to daily MVPA. However, it is unlikely that levels of activity differed systematically between these two time periods.

Interpretation

The results from these two papers illustrate some common themes. The first paper indicated that few of the sociodemographic characteristics investigated were associated with both recorded and reported measures of daily time spent in MVPA, and the second paper showed that it was common for people’s activity to be ranked differently according to the measurement method. These findings probably reflect differences in the way activity is ascertained by questionnaires and activity monitors. Activity of ‘moderate to vigorous’ intensity covers a very wide range of behaviours, such as cycling, swimming and team sports, which are undertaken in a variety of settings and may be differentially related to sociodemographic, health and contextual characteristics. Questionnaires collect information on physical activity behaviours, while devices measure physical movement. Although our experience, as well as that of others, suggests that pedometers or accelerometers capture walking adequately for most purposes, other devices or, indeed, self-reported measures may be required to capture a wider range of activities. At the same time, the different specificities of measurement techniques need to be weighed against the practical implications of using them in different situations. In some cases, it may be helpful to use a combination of reported and recorded measures to establish a more accurate and comprehensive assessment of context-specific behaviours.

Rationale

Previous research suggests that the physical environment is an important influence on active commuting, and the distance of the daily commuting journey has been shown to have a strong and consistent influence on travel behaviour. 79 Distance between home and work can be assessed in several ways: by asking people to report the length of the route, by modelling a likely route or by measuring the length of the actual route. Although distance to work may be relatively accurately recalled because the journey to work is made regularly, self-reported distance may still be subject to recall bias. When modelling routes using a GIS, it is often assumed that the shortest available route is used with the aim of minimising travel time,80 but this may be different from the actual route used, for example if commuters take children to school on the way to work or go shopping on the way home. The choice of route may depend on several factors, including the social context in which household decisions are made as well as individual preferences and physical constraints. One way of capturing data on the actual routes followed by commuters is to use GPS devices. Although these are increasingly small, unobtrusive and affordable, the participant has to remember to wear the device and recharge it each night, and considerable effort is required to clean, process and analyse the resulting data. To explore the implications of relying on modelled routes instead of collecting GPS data from a large number of participants, we investigated the differences between distances estimated on the basis of GIS-modelled routes and those derived from actual routes recorded using GPS. 81

Methods

We derived distances representing the shortest route available between participants’ home and work locations with the help of the route networks created in the GIS. For participants who reported having used the car, we used the car route network to model the shortest possible route from their home to their workplace. For participants who reported having walked or cycled, we used the pedestrian and cyclist route network to model the shortest possible route. Distances estimated on the basis of GIS-modelled routes thus reflected the actual modes of transport used by each participant.

Actual routes were obtained from the GPS receivers worn by a subsample of participants in 2010 and 2011. Briefly, this involved extracting GPS data for the relevant trip in the GIS, as well as any intermediate destinations that were visited en route, with the help of background mapping and aerial photography. Scattering of data points resulting from signal error or reflection from buildings was removed manually. We used multilevel mixed-effects generalised linear regression models to assess the concordance of GIS-modelled and actual commuting routes; in the published paper, we also compared other features of the modelled and actual routes travelled between home and work. 81

Results

We found that the actual routes taken by commuters often diverged from modelled routes, sometimes substantially. On average the actual routes were 27% or 4.3 km longer than modelled routes, but this summary result conceals important differences depending on which mode of transport was used. Actual routes were even longer among those who travelled to work by car, either singly (35% or 6.5 km) or in combination with walking (25% or 6.2 km) or cycling (26% or 5.0 km). In contrast, the lengths of actual routes followed on cycling trips were closer to those of the modelled routes, with an average difference of 23% or 0.9 km. Actual walking trips were 13% or 0.2 km shorter than modelled routes, suggesting that participants who walked to work may have used cut-throughs and other paths not present in the route networks used to model routes. Agreement between route lengths was highest for walking and cycling, with concordance coefficients above 0.93, and lowest for trips made entirely by car, with a concordance coefficient of 0.44.

Taking as an example one participant who travelled 20 km from work to home by car, the shortest route modelled using the GIS underestimated the actual distance travelled by 79% and overlapped the actual route for only 15% of its length. For a second participant who walked 2.2 km from work to home, the route was more similar in length to that modelled using the GIS but only 19.2% of the route overlapped the modelled route. Figure 3 provides a graphical example of this type of divergence between actual and modelled routes.

FIGURE 3.

Comparison of actual and modelled commuting routes. The actual route taken – by bicycle in this example – is longer than the modelled route and overlaps only in parts. In practice the hypothetical participant takes a less direct route, travels along quieter roads and passes fewer destinations than the modelled route would suggest.

Interpretation

Modelled routes and actual routes can differ substantially. Many commuting journeys are not made with the aim of minimising travel time by choosing the shortest distance, but instead often incorporate other destinations. The next chapter offers additional insights into the motivations for route choice, either to avoid undesirable features of the shortest available route or to experience desirable features of the alternative route. Nevertheless, the shortest modelled route to a destination – whether this is a workplace, or the nearest bus stop or access point to the path on the busway – may be a reasonably accurate estimate of the actual distance followed, particularly for walking or cycling trips, and can serve as an indicator of the accessibility of that destination for a given participant (see Chapter 5, Linking environmental change with travel behaviour change in commuters).

Rationale

In the previous section we noted that the actual routes taken by commuters often diverged from the shortest route available, in some cases substantially. This provides a clue that commuting decisions may be more complex than commonly assumed. From the outset, we had envisaged using qualitative methods to explore, and arrive at an ethnographically ‘thick description’ of, individual everyday commuting decisions. 82 The semistructured interviews we conducted with this end in mind tended to revolve around the practicalities of commuting, that is how people made commuting decisions to fit their social lives and any structural barriers that prevented them from making their preferred choices, but it was difficult to get people to reflect on their commuting behaviour beyond these concerns. We therefore decided to adopt ‘photo-elicitation’ techniques – using participant-produced photographs as an interview guide – to gather unstructured, open-ended information on participants’ experiences and the meanings they gave to them. In this section, we describe the ‘photovoice’ procedure we developed to address this challenge and the methodological questions it helped us to address. 83

Methods

Qualitative research that aims to capture experiences that are not easily verbalised usually adopts an ethnographic method, in which the researcher observes and participates in the behaviour or event of interest, and engages the participant in informal conversations about the meanings given to these. The difficulties of getting participants to reflect on their regular commuting behaviour raised the question of how we would be able to obtain observational ethnographic insights into commuting experiences and decision-making without physically joining each participant on their commute. So-called ‘walk-along’ interviews may have been successfully adapted as ‘bus-alongs’ in other studies but seemed less than ideal for cycle commuting, which was prevalent in our sample. Instead, we decided to attempt to capture commuting experiences ethnographically with the help of the visual method of photo-elicitation.

Participants were given low-tech, unintimidating disposable cameras; others chose to use their own digital cameras or mobile phone cameras. With minimal direction, they were asked to take some time on their way to or from work to pause on their journeys and reflect on aspects of their behaviour that they might take for granted or undertake habitually. They could document their actual commuting journeys and aspects or experiences of the journeys that were significant to them, or an ‘ideal’ (desired) commute. We subsequently met participants for follow-up interviews, which lasted between 20 and 60 minutes. These elicitation interviews were purely driven by the discussion of the photographs produced by participants or the subset of photographs that the participants had selected for discussion.

Participants sorted their photographs into a particular order, usually to correspond with the stages of their usual journey. Many had prepared the narrative, or purpose, of their photo story before the interview. Some also provided written ‘memos’ with their photographs, which were also analysed. At the end of the interview, participants drew the journey they had documented on a map; some pointed out where the photographs had been taken along the route. In addition to the photographs and subsequent interviews, which were audio-recorded and transcribed, ethnographic data collection also included general observation of travel practices in the local area and local public discourses around travelling, in order to further contextualise participants’ representations of travel behaviour. 84 We framed our interpretations within anthropological theories of well-being,85 visual anthropology86 and social practice theories,87 eventually funnelling our analytical findings towards three themes. 88

Results

Participants produced over 500 photographs in total. Most strikingly, when applied to commuting the photovoice method brought to light a range of experiences that had been absent in semistructured interviews focusing on the practicalities of commuting. In particular, many of the photovoice accounts focused on positive experiences of what we termed ‘well-being’ in commuting. For example, a number of participants produced accounts of themselves enjoying pleasant scenery. Others appreciated commuting as a relaxing or otherwise valuable transitional time between home and work life. Still others supplied photographs of unorthodox use of the busway, even before its opening (Figure 4). The insights gained by this method are described in more detail in the following chapter.

FIGURE 4.

Participant-produced photographs of well-being in commuting. Photographs reproduced with permission of participants.

Interpretation

Although our approach produced data that were rich and strikingly different from those produced from our other quantitative and qualitative methods, the question then arose as to how these data should be treated analytically. Did these photovoice narratives produce credible insights into the actual experiences, or at least aspirations, of Cambridge commuters, or did the participant-produced photographs simply represent the subjects of relatively easy photo opportunities? We suggest that the photovoice procedure allowed participants to uncover and subsequently describe different, but no less real, experiences and emotions related to commuting. These experiences were captured in photographs and richly described during the elicitation interview. Moreover, many used the opportunity of the elicitation interview to explain how they had failed to capture all of their experiences with their cameras, narrating these instead as ‘would-be’ pictures – still continuing the same themes as they relayed through their photo-stories. This suggests that their photovoice reflected more than simple ‘Kodak moments’ and provided insights into the actual, in part inner, experiences of commuters. Finally, not all participants documented or described positive experiences of commuting, and narratives of stress and distress could also be explored in greater detail with the help of participants’ photographs.

The different accounts produced in the initial and photo-elicitation interviews thus did not discredit each other’s findings, but provided different kinds of insights into commuting experiences, the latter providing insights suitable for a ‘thick description’ even if these might have been difficult to obtain by standard ethnographic methods such as participant observation. We conclude that photography can produce rich observational data in addition to traditional methods of interviewing and focus groups, and thereby help to document and understand a more complete and complex picture of commuting-related experiences. 89–91

Physical activity

Previous research suggests that active commuting is associated with higher levels of physical activity. However, most studies of this relationship have considered only the main mode of transport as the independent variable in their analyses. 23,95–97 Having found that a substantial proportion of commuters used a combination of active and motorised modes of transport, we deduced that using a measure of time spent in active commuting would more accurately capture all relevant activity on the journey to and from work.

Using objective measures of daily time spent in MVPA collected with accelerometers in 2009 (see Chapter 2, Derivation of key variables),98 we found that time spent in active commuting was associated with time spent in total physical activity, even in a sample in which participants recorded a relatively high mean of 55 minutes of daily MVPA (Table 6). Compared with those who reported no walking for commuting, those who reported spending 1–149 minutes per week walking for commuting recorded an additional 6.5 minutes of daily MVPA, and those who reported spending more than 150 minutes per week recorded an additional 14.3 minutes of daily MVPA (p = 0.01). We did not observe statistically significant associations between time spent cycling for commuting and total physical activity in men or women, possibly in part due to the underascertainment of physical activity associated with cycling by hip-worn accelerometers. Total time spent in active commuting was not associated with MVPA in men; but for women, spending 150 minutes or more in active commuting per week was associated with an additional 8.5 minutes of daily MVPA (95% CI 1.8 to 15.3; p = 0.01). The lack of an association in men may have reflected the smaller sample of men available for analysis and their tendency to cycle more and walk less (see Active commuting in context). 62

We also investigated the relationship between active commuting and physical activity using the enhanced activity monitoring data collected in 2010 and 2011, because the combined heart rate and movement sensor data allow for a more accurate assessment of physical activity than hip-worn accelerometers. We analysed 181 commuting trips made by 41 participants, objectively deriving the trips by inspecting the GPS data to identify when participants left home and arrived at work, and using the georeferenced combined heart rate and movement sensor data to estimate the energy expenditure associated with different modes or combinations of modes of transport (see Chapter 2, Derivation of key variables).

Although walking or cycling all the way to work involved substantial energy expenditure as expected, their incorporation into longer motor-vehicle journeys also made an important contribution in this respect. Compared with those who drove all the way to and from work, participants using any other mode or combinations of modes expended significantly more energy per minute (Table 7). Adjusting for other factors, regression analysis showed that those who commuted by bus expended an additional 0.7 METs each minute on average. Combining car use with walking and cycling was associated with additional energy expenditure of approximately 0.6 and 1.6 METs per minute, respectively, while those who walked or cycled all the way to work expended an additional 2.5 and 3.9 METs, respectively.

For trips in which walking and cycling were reported in combination with other modes, the median fraction of the total duration of the trip spent in MVPA was 20%. This equates to around 8 minutes for an average commuting trip of around 40 minutes (see Chapter 3, Validating a self-reported measure of time spent in active commuting), and over the course of 1 week this could amount to over half the recommended weekly levels of physical activity for adults.

In summary, we found evidence of an association between time spent in active commuting and total physical activity, regardless of the main mode of travel to work. Engaging in 150 minutes or more of active commuting per week was associated with achieving an additional 8.5 minutes of daily MVPA among women, which corresponds to about 16% of the achieved daily MVPA in this population and about 30% of the recommended minimum level of MVPA. We further showed that active commuting was associated with substantial additional energy expenditure, and that incorporating even modest quantities of walking and cycling into the journey to and from work could make a substantial contribution to meeting physical activity guidelines.

Health and well-being

Previous quantitative research has found consistent associations between physical activity and physical and mental well-being. However, this evidence comes mainly from studies examining relationships with recreational physical activity. 100–103 Although active commuting may be associated with higher levels of physical activity, an association between active commuting and physical or mental well-being cannot be assumed, and therefore we investigated this association using the summary scores of both physical and mental well-being derived from the SF-8 instrument included in the core questionnaire (see Chapter 2, Derivation of key variables). The analysis was adjusted for known correlates of active commuting (gender, age, educational attainment and distance between home and work) and in particular for time spent in recreational and work-related physical activity (see Chapter 2, Derivation of key variables). Given that being overweight or obese may confound or mediate any relationship between active commuting and well-being, we also tested the effect of excluding weight status from these models.

In this cross-sectional analysis of data collected in 2009, we found that time spent in active commuting was positively associated with self-rated physical well-being and that the largest benefit was associated with participating in at least 45 minutes of active commuting per day. Those with ‘very high’ levels (> 225 minutes per week) of active commuting reported PCS-8 scores that were on average 1.21 points higher than those with ‘very low’ levels (0–29 minutes per week). Including weight status in the analysis slightly attenuated the relationship between active commuting and physical well-being. Although the effect size was lower than the threshold for ‘clinical significance’ in SF-8 summary scores, it may nevertheless correspond to some degree of ‘population significance’ in settings in which active commuting is common.

Although we found no association between active commuting and mental well-being in this quantitative analysis, our qualitative findings suggested that participants articulated well-being in a variety of ways (see Chapter 3, Eliciting motivations and contextual factors). 83 In the photo-elicitation study, many participants produced images of the landscape or of nature, and follow-up interviews contained stories of relaxation and ‘me time’ on the commute. Photovoice showed that people can take pleasure in commuting: ‘I really enjoy my cycle to work . . . [the scenery is] absolutely glorious’ (woman, 34, who commuted by bicycle). Commuters did not necessarily regard travel time as ‘wasted’ time. 50 This well-being in commuting was experienced not only by those who walked or cycled but also by car drivers. One participant, who had spent many years cycling in Cambridge, reflected on the tensions involved in sharing the road with others as a more vulnerable road user and reported: ‘I would say I’m much calmer now I drive and I don’t cycle’ (woman, 57, who commuted by car and parked on-site).

Some photovoice accounts also focused on negative experiences of commuting, such as traffic, stress, danger and boredom. For example, one cyclist photographed a sign that lights up to alert drivers to the fact that they are travelling faster than the speed limit, to document that ‘90% of the cars are breaking the speed limit’ (man, 62, who commuted by bicycle). However, those who depicted negative aspects sometimes also mentioned more positive experiences. Appreciating joyful, positive aspects of commuting may serve as a tactic to make everyday travel routines habitable; however, it seems that a sense of choice is important in experiencing ‘happy’ commutes. 83 For example, one participant reported how he chose a route that he found pleasant: ‘All along my route is . . . it’s all green, there are huge houses with nice gardens and I’ve started to choose the route now for its pleasantness at the start of the morning . . .’ (man, 44, who commuted to by car to a park-and-ride site and continued by bicycle).

In summary, time spent in active commuting was positively associated with a summary measure of physical well-being, independent of recreational and work-related physical activity. It is possible that associations between active commuting and physical well-being would be less favourable in other contexts, for example where active travel may be imposed rather than chosen; in our qualitative analysis, positive experiences of commuting appeared to be associated with having choices. We found no association between time spent in active commuting and mental well-being. However, our qualitative research suggested that well-being during commuting might be experienced and articulated in different ways to those captured in the survey, and not necessarily only in relation to active commuting.

Attitudes and beliefs

Some of the attitudes and beliefs assessed through the psychological measures included in our core questionnaire were associated with time spent in active commuting. For example, we found that participants who reported the highest level of perceived behavioural control with respect to car use – that is, those who agreed that ‘it would be easy’ for them to use a car or that they ‘would be able’ to use a car – were approximately half as likely to report any time spent walking as those who did not. Other findings were not entirely as expected. For example, the most favourable affective attitudes towards car use – that is, broad agreement that ‘it would be pleasant’ or ‘it would be good’ to use a car – were negatively associated with time spent cycling for commuting but positively associated with time reporting walking for commuting. This may in part reflect the prevalence of a pattern of behaviour in which some commuters drove part of the way to work and walked the remainder of the journey, and thus might report both a favourable attitude towards driving and spending time walking to work. 62

The remainder of the psychological measures included in our core questionnaire revealed no associations with the uptake or maintenance of walking, cycling or alternatives to the car in longitudinal analysis, except that participants with less favourable affective attitudes towards car use were five times more likely to maintain an alternative to the car as their most frequently used mode of transport. 56

The limited and sometimes confusing patterns of associations of these psychological constructs with active commuting may in part reflect the facts that the questionnaire items permitted different interpretations and the responses might have been sensitive to unmeasured contextual factors. Our qualitative research illustrated that attitudes relating to commuting behaviour were complex and ambiguous, or involved factors that were outside the scope of our core questionnaire. As we have seen above, participants appreciated having choices about their commute because this gave them control over their overall journey experience. We described the opportunity to control interactions with other commuters and to have personal space on the journey as ‘the choreography of avoidance’. 94 Interviewees often elaborated how they achieved this; it affected all modes of travel, and appeared primarily to be about avoiding unpleasant or stressful interactions with other commuters. Cyclists explained how they chose routes away not only from cars, but also from pedestrians. Others avoided groups of school children on public transport, and still others preferred cars to public transport in order to have their own space, as illustrated by the interview excerpt below. Car and public transport commuters frequently described leaving early to avoid the delay and unpredictability of travelling in rush hour or – less frequently – to find a parking space. 94

I was getting very stressed . . . because of the behaviour of others on the road. Yeah, I think it’s important to have some sort of me-time and I actually find that works very well now I’m in the car for, well, I suppose on average an hour a day.

Woman, 57 years, who commuted by car and parked on site

Finding quick, reliable and convenient commuting routes was described as an important factor in decision-making, but pleasantness or opportunities for relaxation and ‘me time’ were equally valued – often by the same participants. 50

Convenience was a particularly prominent theme. Some participants cited the frequency of public transport or the time when they had to be at work to explain why a particular mode of transport was more convenient. Convenience could trump personal dislike of cycling, and sometimes it appeared to capture the overall burden or ease associated with commuting, as the following interview excerpts illustrate in turn. 51

The buses were really convenient, they would go maybe every 5 minutes.

Woman, 48 years, who commuted by car to a park-and-ride site and continued on foot

I don’t actually enjoy cycling really, I do it purely because it’s convenient.

Woman, 30 years, who commuted by bus and sometimes by car

Because parking is difficult in the centre of Cambridge, on this site in particular, it would be difficult to bring the car in and it’s slower and more expensive so yeah, that’s the main reason . . . Before I lived there I would come in on the motorbike . . . That was quite good but really it’s just more convenient on the train . . .

Man, 55 years, who commuted by motorbike

The complex reasoning behind commuting choices and the importance of convenience were echoed in our mixed-method analyses that explored why participants who reported non-conducive route environments nonetheless walked or cycled to work, and in the experiences and motivations of participants regarding commuting around the period of moving home, which we report below. 51,52

Social factors

Environmental factors

Patterns of busway use

We were able to recruit 1710 eligible busway users for the intercept survey, of whom 178 were unable to stop when approached but later completed the survey via the internet. In total, 57% of participants were intercepted while using the guided bus, 31% were intercepted while cycling on the path and 12% were intercepted while walking on the path.

The characteristics of users intercepted on the busway are summarised in Table 8. In comparison with the main study cohort (see Table 1), the gender composition of the sample of busway users was more balanced (51% male). Fifty-seven per cent of participants described themselves as in full-time employment, and 54% had a university degree or higher educational qualification; although this was higher than the average for Cambridgeshire (36%), it was considerably lower than in the cohort sample. Although the home addresses of the participants were concentrated along and near the ends of the busway, as expected, we intercepted considerable numbers of users who lived in locations more widely scattered over the study area, and 12% lived outside the study area altogether (Figure 6). In other respects, the intercept survey and cohort samples were similar.

FIGURE 6.

Places of residence of users intercepted on the busway. Twelve per cent of users intercepted on the busway lived outside the study area denoted by the circle, some of them outside the border of the map. Contains Ordanance Survey MasterMap^® Integrated Transport Network Layer™ and Address Layer 2 data © Crown Copyright and Database Right 2015.

The sample of busway users revealed statistically significant gender differences in use of the busway, in terms of both mode of transport and trip purpose (Table 9). Men were more likely than women to have been using the path for cycling when intercepted, consistent with our general observation that cycling as a mode of transport was more common for men than for women in and around Cambridge (see Chapter 4, Active commuting in context). Correspondingly, more women than men were intercepted while using the guided bus service. Nearly the same number of men and women were intercepted while walking along the path. Just over half of the users surveyed were travelling to or from work, on business or for study, and men and women were approximately equally likely to use the busway for these purposes. Men were more likely to use the busway for leisure trips, while women were more likely to use it for shopping trips.

Fifty-one per cent of path users reported that they had undertaken moderate physical activity on ≥ 4 of the past 7 days, while 16% reported that they had not engaged in moderate physical activity on any of the past 7 days. In contrast, 29% of guided bus users reported that they had undertaken moderate physical activity on ≥ 4 of the past 7 days, while 31% reported that they had not engaged in physical activity on any of the past 7 days.

The median trip distance reported by busway users was 19.3 km (IQR 9.7–32.2 km), which is more than the average distance from home to work for the participants in our main study cohort (see Chapter 2, Samples). Trip distance varied according to the mode of use of the busway, with median (IQR) trip distances of 24.1 km (16.1–33.8 km), 14.5 km (6.4–27.4 km) and 6.4 km (3.2–17.7 km) for guided bus users, cyclists and walkers, respectively. The median reported trip frequency on the busway in the previous month was 6 (IQR 2–20), which suggested that many participants were using the busway regularly, including for daily commuting.

Perceptions of busway users

Overall, the busway was experienced as having improved transport options and the quality of foot and cycle paths, with weaker overall agreement with statements that it had improved the regularity or reliability of transport services, access to local services or personal safety. Among guided bus users, the most frequently reported benefits were in the range of transport options (83% of participants), the regularity of transport (77%) and the quality of foot and cycle paths (76%). Among path users, the most frequently reported benefits were in the quality of foot and cycle paths (93% of participants), the range of transport options (87%) and personal safety (69%).

A key element of the intercept survey was to capture users’ perceptions of the impact of the busway on their travel behaviour. We asked them what mode of transport they would have used in the absence of the busway for the trip they were currently making, and how the availability of the busway had affected the extent to which they used each of the four main modes of travel.

According to user perceptions, trips on the busway had replaced many car trips, but also some trips that would previously have been walked or cycled, as well as generating some new trips. This pattern of modal substitution differed depending on the mode of transport being used by participants when they were approached for the survey (Table 10). More than 50% of guided bus users indicated that their current trip had replaced a car trip, with much lower proportions reporting that it had replaced a walking (4%) or cycling (3%) trip or that it represented a new trip (8%). Cycle trips were more likely to be described as new trips (15%), and a substantial minority of cycle trips were also said to have replaced car trips (24%).

According to user perceptions, the availability of the guided bus and the path had also affected the overall travel behaviour of the sample. For example, 48% of path users reported that the availability of the path meant that they now cycled more. Substantial minorities of path users also reported that the availability of the bus service meant that they now used the bus more often (43%), cycled more (30%) and used the car less (29%). Similarly, many bus users reported that the availability of guided bus services now meant they used the bus more (58%) and the car less (44%), but bus users also reported that the availability of the path meant that they now walked more (41%), cycled more (41%) and drove less (25%).

The findings of the intercept survey, therefore, indicated a favourable assessment of the busway among many of its users, and provided preliminary quantitative support for the hypothesis that the busway may have influenced travel behaviour in the local population and contributed to increased active commuting or physical activity among some users.

Predictors of busway use

The busway was widely discussed in local media, and walking and cycling on the path began long before the formal opening of the busway and was an important part of the uptake of the intervention. Our core questionnaire data likewise showed that almost all members of our study cohort (more than 99%) were aware of the busway. We went on to study the predictors of the three different modes of use of the busway in the final (2012) phase of our cohort study, with a view to testing the face validity of our measures of intervention exposure (see Chapter 2, Derivation of key variables). 115

People who have recently moved may be more likely to reconsider their commuting behaviour, notwithstanding the results reported earlier (see Chapter 4, Active commuting in context). For this analysis we therefore excluded the 212 members of our cohort who had moved home or workplace at any time between 2009 and 2012. The remaining 453 participants included in this analysis lived on average 6.9 km (SD 8.1 km) from the busway, and their average distance from home to work was 11.4 km (SD 9.5 km). Of these participants, 31% reported having used a guided bus, 40% having cycled on the path and 32% having walked on the path by 2012, albeit not necessarily for commuting.

To account for alternative explanations of busway use in this analysis, we included as covariates the availability of car parking at each participant’s workplace and the shortest distance between each participant’s home and workplace using the pedestrian and cyclist route network, as both of these variables have been shown to influence commuting behaviour (see Chapter 4, Active commuting in context). We found that the availability of the busway made only a small difference to the route lengths and estimated travel times to work for participants. On average, the path shortened the travel distance for participants who had used the path for walking or cycling by approximately 50 m (SD 190 m), and on average the guided bus service reduced the estimated travel time to work for those who had used the guided bus service by just under 1 minute (SD 4 minutes). We found no evidence that these reductions in distance or estimated travel time to work were associated with use of the busway for walking or cycling or for guided bus travel, respectively. Testing the predictive validity of these additional putative exposure measures, while blinded to their relationship with the main travel behaviour-change outcome measures, enabled us to select the most appropriate exposure measure for the subsequent main outcome analyses.

In multivariable logistic regression, we found that residential proximity to the busway was associated with use of the busway for each of the three outcomes studied (Table 11). The strength of the relationship meant that, compared with those living 9 km away, participants living 4 km along the pedestrian and cyclist route network from the nearest guided bus stop were 53% more likely to have used the guided bus service, and those living 4 km from the nearest access point to the path were 34% more likely to have walked along the path and over twice as likely to have cycled on the path.

Few of the other covariates were associated with any of the three outcomes. Women were less likely than men to report having cycled on the path (OR 0.41, 95% CI 0.24 to 0.69). Living in a rural area increased the odds of having used a guided bus service or having cycled on the path by between two and three times compared with living in an urban area. The association between proximity and use of the guided bus service was also strengthened among those who lived in towns, and that for walking on the path was strengthened among those who lived in towns and villages, compared with those who lived in an urban area. This may partly reflect the fact that in rural areas both the population and the transport infrastructure tend to be more clustered (in and around villages).

The association of use with proximity was stronger for cycling than for walking or bus use. One possible explanation is that the strength of the associations reflected the different ways in which the busway and path were used. It has been proposed that the time for which people are willing to travel to access public transport infrastructure increases with the total trip duration. The data from our intercept survey suggest that, on average, guided bus users may have travelled further than cyclists on each trip. Bus users would, therefore, be expected to be less sensitive than cycle users to a given increment in access distance. Walking on the path as captured by our core questionnaire could reflect a mixture of walking to access the bus service and walking in its own right, which may explain why the association of walking with proximity was intermediate between those for cycling and bus use.

These results, therefore, confirmed the face validity of our graded measure of intervention exposure based on the shortest network distance between each participant’s home and the nearest access point to the busway, and that we should use this – rather than the alternative measures of change in distance or travel time to work induced by the busway – as our exposure measure in the subsequent main outcome analyses. The results further show little association between the sociodemographic factors included in our study and use of the busway, apart from whether participants lived in urban or rural contexts and a further confirmation that women were less likely to cycle than men (see Chapter 4, Active commuting in context).

Changes in mode of travel to work

Initially, we used information from the 7-day travel record – in which we ascertained the modes of travel used on each trip to and from work during the past week at baseline in 2009 and at follow-up in 2012 – to study the impact of exposure to the intervention on three outcomes: the proportion of trips that involved any active travel, the proportion of trips that were made entirely by car and the proportion of trips that involved any public transport.

We categorised changes in the first two outcomes in five groups representing a large increase (between 30% and 100%), a small increase (between 0% and 30%), no change, a small decrease (between 0% and 30%) and a large decrease (between 30% and 100%). For example, a participant who previously drove all the way to and from work every day of the week, and now walked to the busway and took a guided bus (and the reverse on the way home) on 1 day each week, would be categorised as having experienced a small (20%) increase in the proportion of trips involving active travel and a small (20%) decrease in the proportion made entirely by car. Another participant who made a similar change on 2 days each week would be categorised as having experienced a large (40%) increase and a large (40%) decrease, respectively. The change in the proportion of trips that involved any public transport was categorised in only three groups representing increase, no change and decrease, to reflect the distribution of this variable, particularly the smaller number of cases in which the proportion had changed.

We took steps to account for a range of alternative explanations of changes in commuting patterns in our analyses. Moving home or workplace may lead not only to a reassessment of commuting options, but also to changes in commute distance, public transport commute travel time or distance from home to the busway, any of which could have prompted changes in commuting patterns. We therefore included continuous variables representing these covariates in our models (see Chapter 2, Derivation of key variables), along with binary variables to account for changes in parking provision at work and the number of children in the household.

The analysis was adjusted sequentially. Initially, we adjusted for distance between home and workplace and for the characteristics of the commute that may change with relocation. Subsequent adjustments included sociodemographic characteristics and the urban–rural context in which participants lived.

We found that a substantial minority of participants had experienced a change in mode share at follow-up (cf. Chapter 4, Patterns of active commuting), with 39% reporting a change in active mode share (with an average change of –1.3%), 37% a change in car mode share (average change +3.4%) and 24% a change in public transport mode share (average change –1.0%). As noted earlier, it is possible that these overall trends, away from active travel and towards the use of the car, reflect the effects of an ageing cohort who tend to move further from work over time (see Chapter 4, Active commuting in context). They contrast with the changes reported in the intercept survey of busway users, who tended to report a shift away from the car and towards more active travel and public transport.

Against this background, we found that living closer to the busway was consistently associated with changes in the commuting outcomes we examined (Table 12). Proximity to the busway was associated with a large increase in active mode share and a large decrease in car mode share. Taking the square root transformation of the exposure variable into account, the results correspond – for example – to participants living 4 km from the nearest access point to the busway being nearly twice as likely to experience a large increase in the proportion of commuting trips that involved any walking or cycling [relative risk ratio (RRR) 1.80, 95% CI 1.27 to 2.55] as those living 9 km away, and approximately twice as likely to experience a large decrease in the proportion of commuting trips made entirely by car (RRR 2.09, 95% CI 1.35 to 3.21). These effect sizes could equally be expressed in terms of any pair of distances representing a one-unit increase in the square root of the distance in km, such as the difference between 1 km and 4 km. Proximity to the busway also protected against a small decrease in active mode share, but it was not associated with changes in public transport mode share (data not shown).

Table 12 shows that most effects of proximity to the busway on mode of travel to work became statistically significant only after adjustment for commute distance and were further strengthened by adjustment for sociodemographic characteristics. The statistical relationship between exposure and behaviour change was, therefore, masked by associations between exposure and some of these covariates, as illustrated for example by the observation that younger and more mobile participants tended to live closer to where the busway was built, as discussed earlier (see Chapter 2, Derivation of key variables). Further adjustment for the relevant baseline commuting pattern of each participant or their car ownership or workplace parking provision did not materially change the results, nor did restricting the analysis to individuals who did not move home or workplace.

We also investigated whether or not the intervention had an effect on the weekly number of commuting trips. This may be important because changes in physical activity and carbon emissions due to the busway would depend not only on changes in the proportion of trips made by different modes of transport, but also on changes in the absolute quantity of travel. However, we found that proximity to the intervention was not associated with a change in the weekly number of commuting trips in our sample.

In summary, against a background of increasing car use and shift away from active travel in our cohort, we found that exposure to the intervention was significantly associated with changes in mode of travel to work after adjustment for multiple potential confounders. Proximity to the busway was associated with a large increase in active travel commute share and a large decrease in car mode share, a pattern consistent in part with the perceptions of busway users found in the intercept survey. The lack of an effect on public transport mode share in the cohort analysis may reflect two factors that reduced statistical power to detect such an effect. The first is that the average change in public transport mode share was much smaller than the average changes in the other mode shares, and the second is the comparatively small proportion of trips (15%) that involved any public transport at baseline.

Changes in time spent in active commuting

We used the same measure of intervention exposure to estimate the effect of the intervention on time spent in active commuting. As noted earlier, this was in the context of a significant decrease in time spent in active commuting in our cohort between 2009 and 2012 (see Chapter 4, Patterns of active commuting). For this analysis changes in time spent in active commuting were categorised in three groups representing increase, no change and decrease, given that a substantial number of participants reported no active commuting at baseline or follow-up. We also conducted a sensitivity analysis using a more stringent definition of change, comparing participants whose active commuting increased or decreased by 50 minutes per week or more with those reporting changes of less than 50 minutes per week (including no change). We also stratified our analysis to compare the effects of the intervention among participants who had reported any active commuting at baseline and those who had not.

As with the previous analyses, we made progressive adjustments to our models to account for potential alternative explanations of changes in active commuting. We adjusted first for age and gender, and then for further sociodemographic characteristics that we had previously found to be associated with active commuting and changes in active commuting (see Chapter 4, Active commuting in context). At this stage we also adjusted for participants’ active commuting at baseline, to account for the possibility that high (or low) levels at the outset might have been associated with particular changes in active commuting at follow-up. We included further adjustments for major changes in life circumstances ascertained at follow-up, in particular whether a participant had moved home or work during the study (see Chapter 2, Samples).

In initial multivariable multinomial regression models adjusted only for age and gender, exposure to the busway was associated with change (both increase and decrease) in total weekly time spent in active commuting and change (both increase and decrease) in time spent cycling for commuting (Table 13). After further adjustment, the effects of exposure to the busway on overall time spent in active commuting ceased to be statistically significant, but the association with an increase in weekly cycle commuting time persisted, and participants living 4 km from the busway were one-third more likely to have increased their cycle commuting time than those living 9 km away (RRR 1.34, 95% CI 1.03 to 1.76). Among those who had increased their cycle commuting time, the increase was often substantial, with a mean increase in this group of 87 minutes per week. The effect of exposure to the busway was somewhat strengthened in the sensitivity analysis restricted to larger changes, in which participants living 4 km from the busway were 44% more likely to have increased their cycle commuting by 50 minutes per week or more than those living 9 km away (RRR 1.44, 95% CI 1.03 to 2.03).

Although there was no evidence of a statistically significant effect of exposure to the intervention on active commuting in the sample as a whole, there was a significant effect among those who did not report any active commuting at baseline. Among this subgroup of participants, those living 4 km from the busway were 76% more likely to have increased their active commuting time than those living 9 km away (RRR 1.76, 95% CI 1.16 to 2.67), whereas there was no effect on overall time spent in active commuting among the subgroups of participants who reported either 1–149 minutes per week or 150 minutes per week or more of active commuting at baseline.

In summary, exposure to the intervention was associated with an increase in time spent cycling to and from work, and with an increase in overall time spent in active commuting among those who reported no active commuting at baseline. We found no evidence of an effect of proximity to the busway on changes in time spent walking on the way to and from work. This may reflect the lower level of walking, relative to cycling, for commuting in our sample, and that changes in walking were reported by fewer participants than changes in cycling (see Chapter 4, Patterns of active commuting).

Changes in time spent in physical activity

We went on to investigate whether or not proximity to the busway was associated with changes in three outcomes reflecting broader impacts on physical activity (see Chapter 2, Derivation of key variables). Our analysis followed the same approach, except that the outcomes were categorised into tertiles (corresponding approximately to an increase, no change and a decrease) because no participants reported exactly the same quantity of physical activity at both time points.

In contrast to the background changes in time spent in active commuting, we found that time spent in recreational and total physical activity changed little in our cohort between 2009 and 2012. The median total time spent in physical activity was 423 minutes per week (IQR 232–675 minutes per week) at baseline and 407 minutes per week (IQR 240–631 minutes per week) at follow-up, and the difference was not statistically significant (p = 0.12). A similar pattern was observed for time spent in recreational physical activity (baseline: median 282 minutes per week, IQR 150–523 minutes per week; follow-up: median 279 minutes per week, IQR 146–480 minutes per week; p = 0.28).

In multivariable multinomial regression models, we found no evidence of a significant effect of proximity to the busway on the time spent in recreational or total physical activity. In the minimally adjusted model, proximity to the busway was associated with both an increase and a decrease in time spent cycling for commuting and recreation combined (shown in the table as ‘total cycling’). The association with a decrease in cycling ceased to be statistically significant after further adjustment, but the association with an increase persisted. In the maximally adjusted model, those living 4 km from the busway were 32% more likely to report an increase in time spent cycling for commuting and recreation combined than those living 9 km from the busway (RRR 1.32, 95% CI 1.04 to 1.68). No such association was observed for ‘total walking’ or ‘total walking and cycling’ (Table 14).

In summary, we found that exposure to the intervention was associated with an increase in time spent cycling for commuting and recreation combined, with an effect size very similar to that for time spent in cycling for commuting alone. We found no evidence of an association with an increase in recreational or total physical activity.

Changes in carbon emissions attributable to commuting

While the busway had complex effects on commuting patterns, exposure to the intervention was associated with a shift away from the car and towards active modes of travel for the journey to and from work and a significantly greater likelihood of an increase in weekly time spent cycling. We went on to assess the impact of the busway on carbon emissions, using the same general approach as in the preceding outcome analyses. Specifically, we investigated whether proximity to the busway was associated with an increase or a decrease in carbon emissions attributable to commuting trips in our cohort.

We found that, for commuting trips made by public transport and not involving use of a car, the mean estimated carbon emissions were 1.38 kg per trip in 2009 and 1.78 kg per trip in 2012. For commuting trips involving any use of a car (with or without public transport), the mean of the estimated carbon emissions ranged from 2.70 kg per trip for car-only trips in 2009 to 3.32 kg per trip for trips made by car plus public transport in 2012. We therefore defined participants with a change in their estimated carbon emissions of < 3 kg per week – approximating to the emissions attributable to one ‘average’ one-way car-based commuting trip in our sample – as having made only a small change, and included them with participants who reported no use of motor vehicles and thus no carbon emissions at either time point in our reference outcome category of ‘no change’.

In unadjusted multinomial models, we found that proximity to the busway was associated with a reduced likelihood of both a decrease and an increase in carbon emissions over time. This pattern remained when the models were adjusted for age and gender. After further adjustment for all other covariates, however, the counterintuitive association between proximity and a reduced likelihood of a decrease in carbon emissions ceased to be statistically significant, and proximity to the busway remained associated only with a reduced likelihood of an increase in carbon emissions (RRR 0.57, 95% CI 0.43 to 0.75). In other words, for those living 4 km from the busway, the likelihood of an increase in commuting carbon emissions was approximately half that for those living 9 km from the busway. The other covariates associated with either a decrease or an increase in carbon emissions in maximally adjusted models were education, car ownership, baseline carbon emissions, urban–rural status and moving, all with effects in the expected direction. We further found that the effect estimate for the intervention was only minimally affected by removing workplace car parking, urban–rural status or baseline carbon emissions from the models. Excluding those who had moved home from the analysis led to a strengthening of the effect (albeit with a wider CI owing to the smaller sample size: RRR 0.44, 95% CI 0.26 to 0.73), which may reflect the fact that moving home tended to be associated with a lengthening of the commuting journey in our cohort.

Investigating causal mechanisms

Having estimated the size of the effect of the busway on commuter travel behaviour, we went on to investigate the mechanisms by which the intervention may have led to behaviour change. Identifying plausible mechanisms in this way can increase understanding of an intervention and provide further support for causal inference. 118 As we found no evidence of a significant effect on walking, this analysis was focused on elucidating mechanisms leading to an increase or decrease in time spent cycling for commuting.

We identified three groups of mediating factors that could be involved in this causal pathway: self-reported use of the path along the busway, changes in perceptions of the route environment between home and work and changes in attitudes and beliefs relating to car use (see Chapter 2, Derivation of key variables). We further hypothesised that the most likely mechanisms would be those that included only significant associations between variables (p < 0.05 when adjusted for covariates but not the other mediators). On this basis, we first calculated all statistically significant associations between the various exposure, mediator and outcome variables. We then reduced the complexity of the model by excluding any mediators that were associated only with variables from within the same group of mediators (e.g. perceptions of the route environment), as well mediators that were not on a pathway between exposure and outcome, and eliminated mechanisms that contained more than two mediators from within the same group of mediators.

The resulting model identified one direct association between proximity to the busway and an increase in time spent cycling, as well as several indirect pathways (Figure 7). The direct association was equivalent to that reported in the previous section. The model further showed that although there was no direct association between exposure and a decrease in time spent cycling, there were two indirect pathways leading to this outcome.

FIGURE 7.

Mediators of changes in time spent in cycle commuting. This is a reduced model showing only those associations significant at p < 0.05. The dashed line is unidirectional towards more cycling.

We then used path regression analysis to test which of these pathways were statistically significant overall. We found that the indirect pathways via use of the path leading to either an increase or a decrease in cycle commuting time were statistically significant and explained most of the effect of the intervention. The indirect pathway towards an increase in cycling explained 86% of the overall effect, while the direct pathway between exposure and outcome explained only 13% of the effect. None of the other pathways towards an increase in cycle commuting time via perceptions of the route environment was statistically significant. Similarly, the indirect pathway towards a decrease in cycling explained 96% of the overall effect, while the indirect pathway via convenient public transport was not statistically significant.

We further divided our sample and repeated the analysis separately for those reporting (a) no active commuting and (b) any active commuting at baseline. This led to the identification of six possible pathways linking proximity to the busway to changes in cycling. In the group reporting no active commuting at baseline, only the indirect pathway linking proximity with an increase in cycling via use of the path was statistically significant, and this explained 80% of the overall effect. Among those who already reported some active commuting at baseline, proximity to the busway was associated both with an increase and a decrease in cycling, and only the indirect pathways via use of the path were statistically significant.

These results showed that of a number of plausible mechanisms tested, self-reported use of the path explained the majority of the association between proximity to the busway and an increase in time spent cycling for commuting, while changes in perceptions of the environment or changes in Theory of Planned Behaviour constructs were not significant in mediating the observed changes in commuting behaviour. It should be noted that that if the questionnaire items used to ascertain these cognitive constructs had been framed differently, the analysis might have suggested stronger support for their role as mediators. Nevertheless, showing that the effect of proximity to the busway on commuting behaviour was largely explained by use of the busway provides further evidence against the possibility that the effects observed in the main outcome analyses could have been attributable to confounding, and indicates a plausible mechanism by which the busway may have changed commuting behaviour: where people live near new infrastructure that supports active commuting and begin to use it (for any purpose), this may lead to changes in their overall commuting behaviour. Contextual factors may have been important in bringing about such an outcome. In particular, the intervention we studied was implemented in a context in which cycle commuting was already common and compatible with high social status, while car commuting was comparatively burdensome owing to congestion and difficulties with parking. All of these factors might have encouraged participants to trial the path along the busway and then to use it more consistently for cycling. In addition, these results suggest that the association between proximity to the pathway and decreases in weekly cycle commuting time in the minimally adjusted models reported earlier represented a real effect of the busway among some people, particularly those with higher levels of active commuting at baseline.

Changes in time spent in physical activity

The distribution of time spent in active commuting was skewed at baseline and follow-up, as discussed earlier (see Chapter 4, Patterns of active commuting). In our analysis of longitudinal associations between active commuting and self-reported physical activity in the sample used for the main outcome analyses reported above, we therefore categorised exposure to active commuting as an increase, no change or a decrease in weekly time spent walking, cycling or in active commuting over time, and examined associations with changes in physical activity over the same time period categorised as tertiles. The highest tertile represented a large increase (mean +390 minutes per week for total physical activity and +323 minutes per week for recreational physical activity), and the lowest tertile represented a large decrease in physical activity (mean –495 minutes per week for total physical activity and –370 minutes per week for recreational physical activity). Covariates were added to the analysis in steps and the final model was adjusted for age, gender, car ownership, access to a bicycle, distance from home to work, baseline BMI and baseline total physical activity.

In maximally adjusted multinomial regression models, a decrease in active commuting was associated with a greater likelihood of a large decrease in total physical activity (RRR 2.1, 95% CI 1.1 to 4.1). In other words, participants who reported a decrease in active commuting between 2009 and 2012 were approximately twice as likely to report a large decrease in their total physical activity as those who reported an increase in active commuting. Conversely, an increase in active commuting was associated with a greater likelihood of a large increase in total physical activity (RRR 1.8, 95% CI 1.0 to 3.4), although this result was of borderline statistical significance (p = 0.06).

We further found that an increase in cycle commuting time was associated with a greater likelihood of both a large increase (RRR 2.5, 95% CI 1.2 to 5.0) and a large decrease (RRR 3.0, 95% CI 1.4 to 6.3) in total physical activity. A post-hoc t-test indicated that participants whose cycle commuting increased and total physical activity decreased reported significantly higher baseline physical activity than those whose commuting and total physical activity both increased (930 vs. 471 minutes per week; p = 0.001). Finally, we found no associations between changes in time spent walking for commuting and total physical activity, or between any of the active commuting exposures and self-reported recreational physical activity.

Changes in health and well-being

We also investigated whether or not walking and cycling for commuting predicted sickness absence from work, physical and mental well-being, and BMI at follow-up.

We analysed these longitudinal associations separately for time spent walking for commuting and time spent cycling for commuting. We categorised each of these exposures in respect of whether it was reported at all at both baseline and 1-year follow-up (yes or no) and whether or not it had changed over that time period (increased, decreased or not changed). We then tested whether or not maintenance of these behaviours predicted each of the four health indicators ascertained at 1-year follow-up, and if changes in the behaviours predicted changes in each of the four health indicators over the same time period. We also adjusted each analysis for the value of the respective outcome indicator at baseline. To increase our sample size, and because this analysis was not focused on the effects of the busway, we pooled data across all three 1-year transitions available in the study, that is those between 2009 and 2010, between 2010 and 2011 and between 2011 and 2012. Each participant was included only once in each analysis, based on the first year of the study in which they had returned a questionnaire with a completed travel-to-work record. The regression analyses were adjusted for potential confounders, as detailed below.

We found that maintenance of cycle commuting predicted three of the four health indicators 1 year later (Table 15). On average, those who reported maintaining cycle commuting reported lower sickness absence, higher mental well-being and lower BMI at 1-year follow-up than those who reported no cycle commuting. The effect size was equivalent to 1.2 (p = 0.005) or 1.1 (adjusted for baseline sickness absence; p = 0.004) fewer days of sickness absence over 1 year. This compares with a mean of 3.3 self-reported sick days per year (SD 11.1 days) in the cohort. The effect sizes for physical and mental well-being were of a similar order of magnitude as those we found in our earlier cross-sectional analysis (see Chapter 4, Relationships with physical activity, health and well-being), although the findings for physical well-being were non-significant and the findings for mental well-being became non-significant after adjustment for baseline mental well-being. The difference in BMI was also attenuated and became non-significant after adjustment for baseline BMI.

Changes in active commuting were associated with changes in the health indicators in two ways. First, an increase in time spent cycling for commuting was associated with a net increase of approximately one unit in the PCS-8 score after adjustment for potential confounders and baseline PCS-8. Second, an increase in time spent walking to work predicted a net decrease in BMI of 0.2 kg/m², with a reduction in time spent walking to work predicting a net increase of 0.2 kg/m². Although not statistically significant, these associations are consistent with the hypothesis that active commuting has positive effects on health and well-being.

In summary, changes in active commuting were associated with commensurate changes in total self-reported physical activity over a 3-year period of observation and we found no evidence of a compensatory change in self-reported recreational physical activity. We further found that active commuting was associated with lower sickness absence at work and improved indicators of health and well-being in some analyses, albeit with modest effects at the individual level.

Understanding active commuting

Despite 85% of the sample having access to a car, active commuting was a highly prevalent behaviour in our cohort, reflecting the background prevalence of cycling in Cambridge and the selection of largely healthy workers into the cohort. Although time spent in physical activity overall changed little over the duration of the study, time spent in active commuting declined and this was mostly explained by a decrease in time spent cycling. Commuters reported using a great variety of modes and combinations of modes of transport on the journey to and from work, and walking and cycling were often incorporated into longer journeys by car or public transport. On average, 20% of the duration of these multimodal trips was spent above the moderate physical activity intensity threshold, indicating their substantial potential contribution to commuters’ weekly volume of physical activity. Those who spent more time in active commuting were more likely to record higher levels of overall physical activity as measured by accelerometer, and changes in active commuting were associated with commensurate changes in overall physical activity over a 3-year period of observation. Cycle commuting in particular was associated with lower sickness-related absence at work and improved well-being indicators 1 year later, albeit with modest effects at the individual level, and qualitative narratives also indicated experiences of well-being related to active commuting.

Understanding commuting as a complex behaviour led us to explore how it was shaped by complex motivations and circumstances, with a particular focus on social and environmental factors. The convenience, pleasantness, speed, reliability and safety of the transport options available were important considerations in making commuting choices, but these choices were often not straightforward. For example, cycling might have been chosen as the fastest mode of transport, but detours could be incorporated to make a given journey safer and more pleasant away from traffic. Similarly, commuters could have positive attitudes towards one mode of transport, such as the car, while also using another, such as walking or cycling. Commuters tactically accommodated their commute within a broader set of social and physical constraints and contexts. Commuting practices differed between men and women and were shaped by relationships and responsibilities within and beyond the immediate household, for example the need to care for children or elderly relatives. The ability to live closer to work or to be flexible about the timing of the commute was related to higher socioeconomic position, and changing life circumstances such as becoming parents or home owners often required commuters to accept longer journeys from more affordable residential locations outside the city. Although distance between home and work was an important influence on commuting behaviour, it did not necessarily present an obstacle to incorporating walking or cycling into longer journeys made by car or public transport. The likelihood of active commuting was also related to the availability of parking at work, safe, convenient routes for walking and cycling, and convenient public transport.

Evaluating new transport infrastructure

Our evaluation of the busway highlighted the complexity of the intervention in providing a new guided bus service as well as a traffic-free path for walking and cycling. While the guided bus service provided an obvious alternative to car travel, it also had the potential to replace trips previously made by walking or cycling and to generate new trips. Our survey of busway users found that more than half of guided bus trips and one-quarter of cycle trips intercepted on the busway were said to have replaced a car trip, and nearly half of users reported that they now walked or cycled more than they had before the opening of the busway. In cross-sectional analysis of our cohort data, exposure to the intervention – defined in terms of proximity, modelled as the negative square root of the shortest network distance from home to busway – was associated with a greater likelihood of using the busway for walking, cycling and bus travel. Of these, cycling was the mode of transport most sensitive to proximity. The effect of proximity was strengthened in towns for bus use, and in towns and villages for walking, compared with urban areas. Men were more likely than women to have cycled on the busway, whereas individual socioeconomic characteristics did not predict guided bus use or walking.

Against the background of a general decline in active commuting in our sample over time, and using a robust quasi-experimental study design, we went on to show that the provision of the busway was associated with a shift away from using the car and an increase in time spent in physical activity on the journey to and from work. Longitudinal analysis of our cohort data found that exposure to the intervention was associated with nearly a doubling of the likelihood of a large increase in the proportion of commuting trips involving any active travel, and of a large decrease in the proportion of trips made entirely by car, as well as with a correspondingly lower likelihood of an increase in estimated carbon emissions attributable to commuting. Exposure to the intervention was also associated with a greater likelihood of an increase in weekly time spent cycling on the journey to and from work, to the effect that participants living 4 km from the busway were one-third more likely to have increased their cycle commuting time between 2009 and 2012 than those living 9 km away. This effect was maintained in analysis of time spent cycling for commuting and recreation combined, suggesting that an increase in commuter cycling was not offset by a compensatory decrease in recreational cycling, but we found no effect on time spent in physical activity overall. Use of the busway explained the large majority of the observed effect on active commuting, while changes in cognitions relating to the route environment or to car use were not significant in mediating the observed effect. The effect was moderated by baseline commuting behaviour, with a significant effect on overall time spent in active commuting among those who reported no active commuting at baseline. Interviews suggested that although people were unlikely to use the new infrastructure unless it closely matched the journeys they needed to make, a range of other factors informed travel behaviour, and these were dependent on the value attributed to different aspects of the journey experience. These generally involved considerations of comfort, ambience or pleasantness and of feeling safe, which could trump considerations of reliability and speed. Although experiences of the busway were complex and ambiguous, they culminated in meaningful travel behaviour change for some users, through shifts in the balance between influential factors and planning, trialling and adopting new practices over time.

Understanding active commuting

Linking environmental change with travel and physical activity behaviour change

In reflecting on the meaning of our evaluative findings, it is important to bear in mind two aspects of the way in which the study was conceptualised. The first is that we did not set out to evaluate whether the busway was ‘effective’ in any overall or comprehensive sense, and certainly not to evaluate its value for money. While similar transport projects exist around the world and more are planned, a guided busway is only one of many specific ways in which central or local government might seek to improve infrastructure to promote the use of more sustainable modes of transport. Instead, we took the opportunity presented by this natural experiment to investigate a more specific set of research questions focused on linking environmental change with travel and physical activity behaviour change in commuters. The second aspect is that we were not able to conduct this research using a straightforward parallel-group study design, for the reasons outlined in previous chapters. We therefore sought to build an evidential case for causal inference using multiple sources of data and types of analysis. With these considerations in mind, in this section we discuss the meaning of our findings as they relate to the complementary scientific goals of causal estimation, causal explanation and generalisable causal inference. 108

Methodological investigation and development

Although methodological research was not the primary aim of this study, we have made a number of contributions to methods for public health research in the course of its execution. Most importantly, we devised and subsequently validated a self-reported estimate of weekly time spent in active commuting. In doing so, we not only provided the first short, validated measure of this kind in a field that has tended to rely on measures that are either not specific or not validated, but also made a methodological contribution to the challenge of inferring activity patterns from georeferenced activity data, which may play an increasingly important role in future public health studies linking environment and behaviour. Alongside our investigation of different quantitative methods of capturing key constructs of importance to the study, we also explored the qualitative method of photo-elicitation as a means of gathering ethnographic observational data. This resulted in a change in participants’ narratives from more operational considerations of their transport options to more emotional and indirect reflections, for example in respect of commuting as a source of well-being. Perhaps the most important implication of this work is to highlight a more general observation that different elements of the study contributed to our understanding of the research topic in different ways – ranging from the testing of pre-specified statistical hypotheses about aggregate effects of the intervention, to the unstructured and flexible exploration of the detail of what the practice of ‘active commuting’ entails and means for different people.

Challenges of the study

We faced a number of challenges in the design and execution of this study. We have chosen to highlight four of these here because they illustrate some of the difficulties inherent in addressing the research recommendations on physical activity and the environment published by NICE in 2008,17 which formed part of the background to the inception of the study. The first challenge was that of combining different disciplinary perspectives on the assessment of behaviour change in the context of the evaluation of a transport infrastructure project. The fields of physical activity research and transport research have tended to assess travel behaviour using different instruments and summary measures, and we aimed to establish some common methodological ground between these different approaches. This was important in order to be able to assess the effect of the intervention using a suite of complementary metrics to serve the needs of different groups of users of the findings, while also understanding the social and physical contexts of the target behaviours. The second challenge was that of implementing rapid and robust baseline measurement in the somewhat unpredictable setting of a natural experimental study of an intervention that was outside the control of the research team. This required us to make trade-offs between the competing merits of different approaches and technologies for capturing travel and physical activity behaviour change. The third challenge was that of finding a way of defining exposure to the intervention that would enable us to make controlled comparisons and thereby strengthen the basis for causal estimation of the effect of the intervention. This required us to deal rigorously and transparently with the emerging realisation that a simple parallel-group design was not suitable for the evaluation of the busway. The fourth challenge was that of designing and adapting an analytical strategy that was able to cope with the delayed implementation of the intervention. This required us to use different types of data and approaches to analysis to piece together a combination of evidence for causal estimation and evidence for causal explanation along an extended putative causal pathway linking environmental change with behaviour change and population health impacts. In the following sections, we briefly discuss our responses to these four challenges in terms of their implications for the strengths and limitations of the study and for future research, most which have been described elsewhere in the report but are summarised here for the sake of completeness.

Strengths of the study

Together with our funders, we demonstrated a flexible approach to the unpredictable realities of natural experimental research in two important ways: first by implementing rapid baseline data collection in what we believed to be a brief window of opportunity before the originally scheduled opening of the busway, and later by extending the study and endeavouring to make the best possible use of the available data for causal inference and understanding. We measured physical activity in general, and travel to and from work in particular, using a combination of simpler and more detailed, self-reported and objective measures. This enabled us to assess the main outcomes of the evaluation using a comparatively simple self-reported instrument that was readily administered to the entire study cohort in the core questionnaire at baseline – and could be used in future studies – while later adding more detailed measurement data to validate our primary outcome measure and investigate more detailed patterns and relationships. By combining disciplinary perspectives on the measurement of travel to and from work, we were able to collect disaggregated data at the level of the trip and, to some extent, the stages of each trip. This enabled us to identify more complex patterns of behaviour such as the use of different combinations of modes of transport or different ‘intensities’ of car commuting, and, therefore, to investigate their relationships with potential targets of interventions such as the provision and cost of parking at work. We also demonstrated a variety of ways of using qualitative data, separately and in combination with quantitative data, to understand travel behaviour and behaviour change in commuters. For example, we used mixed-method approaches to interrogate initially counterintuitive quantitative data with insights from qualitative data; and we used a variety of qualitative methods, including vignettes constructed from the results of quantitative analyses, to investigate the effects of the busway. Our detailed exploration of the target behaviours and their relationship with perceived and objective attributes of the environment – particularly in relation to the route to work – generated valuable original observational evidence in its own right as well as highlighting the spatial distribution of the target behaviours with respect to the busway, which had an important bearing on our ultimate analytical strategy for the evaluation. We built on the methods of analysis developed in this work to create a general graded measure of intervention exposure with demonstrable face validity that could be adapted to the causal analysis of different outcomes, and an example of how similar principles could be applied in future studies. In our ultimate estimation of the effects of exposure to the intervention, we systematically accounted for a set of confounders representing plausible alternative explanations, thereby providing a considerable increase in rigour over most previous evaluation studies in the field. 17,18,132 Finally, the study provides a rich observational data set that constitutes a valuable platform for further analyses, particularly using the intermediate years of the core questionnaire data and the objective measurement data, about which potential collaborators can find out more at our data sharing portal at http://epi-meta.medschl.cam.ac.uk.

Limitations of the study

Our rapid and rolling approach to the recruitment of participants in multiple workplaces meant that we had no defined sampling frame such as a population register, and, therefore, could not compute a population response rate as such. As is often the case with studies of this kind, our cohort sample included a predominance of women and people educated to degree level or above; although this was partly offset by the more heterogeneous intercept survey sample, it limited the statistical power of the study to detect associations in men and to examine socioeconomic gradients in the effects of the intervention. Our cohort also suffered considerable attrition over time; although this was not surprising, particularly in a city such as Cambridge with a high turnover of professionals, together with the need to extend the study this resulted in a loss of statistical power for comparisons between the first and last years of the cohort study. Although we validated our primary outcome measure, it was far from perfect, and self-reported outcome data are inevitably susceptible to potential bias as a result of intentional or unintentional misreporting. The multiplicity of ways of summarising travel behaviour change – and the distributions of these outcome variables, which rendered them unsuitable for linear modelling – meant that our evaluative findings were expressed in terms of the likelihood of various categorical outcomes, rather than in the neater terms of a simple ‘effect size’ such as an increase in the number of minutes spent cycling per week. We ultimately defined exposure to the intervention solely in terms of where participants lived. Using more detailed individual measures, for example including both the location of the workplace and the route followed between home and work, might have captured exposure more fully and produced stronger associations with behaviour change, but it might also have involved considerably more work for little additional causal understanding. 14,131,141 Our analysis of mediators of the effects of the intervention was limited, both in terms of the types of mediators considered and in terms of how those were measured. It is possible that if the questionnaire items used to ascertain the cognitive constructs had been framed differently, the analysis would have suggested stronger support for their role as mediators. Finally, owing to circumstances beyond our control we were unable to assess the maintenance of travel behaviour change after the intervention over a second year of follow-up as originally planned; and because we did not observe an effect of the busway on overall physical activity after 1 year, we did not attempt to model the direct impacts of the busway on the other health outcomes of interest, but instead investigated their longitudinal associations with active commuting by way of contributing a different piece of the ‘evidence jigsaw’.

Wider contributions

Adam Bostanci (scientific editor, Medical Research Council Epidemiology Unit and CEDAR) led the scientific and technical editing of the content of the final report.

This report represents an original summary and synthesis of a large body of research, most of which has already been published – or submitted or prepared for publication – in more detail in other open-access academic journals. Lena Alexander, Andrew Carse, Silvia Costa, Alice Dalton, Louise Foley, Anna Goodman, Eva Heinen, David Humphreys, Caroline Jones, Joanna Kesten, Calum Mattocks, Oliver Mytton, Richard Prins, Mark Tully and Lin Yang each led one or more of these analyses or publications, and Søren Brage, Simon Cohn, Cheryl Chapman, Carol Desouza, Amelia Harshfield, Natalia Jones and James Smith each contributed to one or more of these analyses or publications.

Baseline data collection was funded by CEDAR, a Public Health Research Centre of Excellence supported by the British Heart Foundation, Economic and Social Research Council, Medical Research Council, National Institute for Health Research and Wellcome Trust under the auspices of the UK Clinical Research Collaboration.

The authors also thank the study participants, many of whom contributed to multiple parts of the study in multiple years; their employers, whose co-operation and support enabled us to recruit participants through workplaces; the Cambridgeshire Travel for Work Partnership, for assistance with recruiting employers; Cambridgeshire County Council, whose co-operation, information and support made it possible for us to use the busway as a research laboratory; CTS Traffic and Transportation, for conducting the intercept survey; all staff from the Medical Research Council Epidemiology Unit Functional Group Team, in particular for study co-ordination and data collection (led by Cheryl Chapman, Fiona Whittle and Andrew Dymond), physical activity data processing (Kate Westgate and Stephanie Mayle) and data management (Lena Alexander, Ruhul Amin, Rupesh Ghelani, Amelia Harshfield and Wing Wong); Natalia Jones and Louise Foley, for their contributions to the collection of interview data and the preparation of survey data for analysis respectively; Stephen Sharp, for statistical advice; Mark Petticrew, Ashley Cooper and Harry Rutter, for their guidance and support as members of the independent study steering committee; and Oliver Francis, for his support with knowledge translation.